Glossary/Benchmark/Percentile scoring

Percentile scoring

Percentile scoring reports the value below which a given share of measurements fall, exposing worst-case behaviour that an average conceals.

Also known as percentile, p99, percentile ranking

Measured by SystemCheck

Part of the four-stage run: a 60-second sustained CPU load reporting throughput, decay and consistency; a memory stage reporting bandwidth and latency; a calibrated GPU raster load reporting frame time, pacing and decay; and a path tracer reporting samples per second. The paragraphs below say which metric carries this term.

The 99th-percentile frame time is the value 99% of frames beat. Reporting it alongside the mean turns a single number into a description of a distribution. You learn not just how fast the typical frame was but how bad the bad ones got.

This matters because perceived quality is driven by the tail, not the centre. Users notice the stutter, the hitch, the moment the machine paused; they do not notice the thousands of frames that were fine. An average weights every frame equally and therefore misrepresents the experience.

Percentile scoring is also more robust than min/max, which can be thrown off by a single outlier from an OS interrupt. SystemCheck builds its consistency figures from the distribution of measured intervals for this reason, so a bad result reflects a pattern of poor delivery rather than one unlucky frame.

Related terms

1% lowThe 1% low is the frame rate of the slowest 1% of frames, capturing the stutters that an average frame rate hides entirely.Read more Consistency scoreA consistency score summarises how evenly a system delivers performance across a run, separating steady machines from ones that are fast on average but erratic.Read more Frame timeFrame time is the milliseconds taken to produce one frame, and it describes smoothness far more accurately than an averaged frames-per-second figure.Read more Run-to-run varianceRun-to-run variance is the spread in results when the same benchmark is repeated unchanged, and it sets the smallest difference you can trust.Read more Score normalizationScore normalization converts raw measurements onto a common scale against a fixed reference, so results stay comparable across machines and runs.Read more

Run the free test All terms